Sie sind auf Seite 1von 61

Mathematical Models for

Earthquake Simulations: Forward


and Inverse Model Applications
on North Anatolian Fault Zone
December 1, 2015

Ayegl Askan

Department of Civil Engineering


and
1
Earthquake Engineering Research Center, METU,

Introduction: Ground Motion Modeling


Three main issues (by Seismo. Soc. of Amer. 1906):
1. the physical earthquake event itself (when,
where, how)
2. the associated ground motions
3. the effect on the structures
a(t)
a(t)

td

Source

Site
Effects

Softer Soil Layers

Bedrock

u0

Structures

Structur
al
Respons
e

..
Path
Effects
2

Definitions: Forward and Inverse Ground


Motion Simulations
Forward Ground
Motion Modeling

a(t)

Structures

a(t)

td

Source

Site
Effects

Softer Soil Layers

.
.

Bedrock

u0

Structural
Response

Inverse Ground
Motion Modeling

Amplificatio
ns

Path
Effect
s

source
inversion
material
inversion

Forward and Inverse Ground Motion


Modeling in Large Basins
Problem: modeling and forecasting ground
motion in large basins with heterogeneous
material properties
Significance: design and assessment process in
hazard mitigation
Overall objectives:
to develop the capability for generating
realistic broadband ground motion using
improved models of complex basin geology and
earthquake sources (Why inversion?)
to use that capability to model and forecast
strong ground motions

Forward and Inverse Ground Motion


Modeling in Large Basins
Challenges:
irregular geometry
heterogeneity and uncertainty in geological and
source parameters
spatial variation of soil properties requires highresolution models and solutions
different natures of high and low frequency
motions
Tools: Supercomputers along with various
numerical techniques, material databases
5

Contents
Inverse models for velocity structures (I.
Theoretical, II. Field work)
Forward models (Deterministic and
Stochastic)
Limitations of these techniques
Applications in Northwestern Turkey:
Validation of ground motion estimations with
a major past earthquake (1999 Duzce eq.)
Laquila earthquake
Practical use of synthetic ground motions
6

Previous Work

(Akcelik et al. 2003)

Elastic 3D waveform inversion


Need to be improved for the anelastic losses in soil media
A new 2D model including anelasticity
Faster optimizer!
Intrinsic attenuation (damping) in soil media is represented
using mechanical analogs and quality factor [ Q=1/2z at
resonance frequency]
7

Forward Viscoelastic Wave Propagation


Problem
2u
2 [ (u v)] f
t

v
u
v
t
t

in x [0,T]
on FS x [0,T]

(u v) n 0

(u v) n
u
(0) u (0) v(0) 0
t

in x [0,T]

u
t

on AB x [0,T]
in

u and v: the state variables,


: the shear modulus,
: the relaxation frequency
of the SLS,
: the spring constant of
the SLS,
f: the source function.

Soln. Technique: Finite Element in space and Finite Difference in


time

min( wavelength) VS min / f max

6
6

1
(2 ) f max SLS

Least squares parameter estimation formulation of inverse


anelastic 2D scalar wave propagation (with regularization)
receivers
inversion field

1 NR
*
2
[
u

u
]
( x x j )ddt
min
u ,v , 2 j 1
0

Subject to:

data misfit

u
2 [ (u v] f
t
v
u
v
t
t
(u v) n 0

on FS x [0,T]

(u v) n

on AB x [0,T]

displacements

u
( 0) u ( 0) 0
t

target

1/2

( )d
2

flat minimum

u
t

in x [0,T]
in x [0,T]
forward
seismic wave
propagation
model

v ( 0) 0

local minimum
(Tikhonov reg.)

Tikhonov reg.
+ multiscale

Total var reg


+ multiscale

Synthetic Examples from Southern


California
Depth (km)

Homogeneou
1x1
2x2
(Askan et al.,
2007; Askan
and
Bielak,
2008;
Askan
et al.3500
2010)
s
3500
3500
0
0
0
3000

2500

10

2000
1500

15
20

1000
0

10

Depth (km)

30

4x4

20

500

30 (m/sec)

32 x 32

3500
3000

2500

10

2000
1500

15
20

1000
10

1000
0

10
20
30
Distance (km)

1000

3500

1500

1500

500

2000

15

2000

15

(m/sec)

2500

10

2500

10

20

3000

20

Depth (km)

20

3000

500

(m/sec)

10

20

30

8x8

1500
1000
0

10

20

30

64 x 64

500

2000
1500

15

1000
0

10
20
30
Distance (km)

1000
0

10

20

500
30 (m/sec)

16 x 16

2500

10

2000
1500

15
20

500

(m/sec)

1000
0

10

20

30

500

(m/sec)

3500
3000

2500

10

2000
1500

15
20

3500
3000

Target
3500
2500

10

1500

(m/sec)

3000

20

2000

15
20

3500
2500

10

2000

15

500

3000

2500

10

20

(m/sec)

3000

1000
0

10
20
30
Distance (km)

500

(m/sec)

10

Synthetic Examples from Southern


California
(Askan et al., 2007;
Askan and Bielak,
2008; Askan
et al. 2010)
Homogeneous
1x1
2x2
___ Inverted
___Target

4x4

32x32

8x8

64x64

16x16

Target

11

Limitations:
Real earthquakes generate much more complex
waveforms than simple pulses!
Higher frequencies require BIG meshes, huge
computing times
Regularization causes numerical problems for the
non-synthetic cases
Caltech, USC, Carnegie Mellon
Objective: 3D version that could work linearly for
real data
Needs a lot of work for realistic inversions of
wave velocity!
12

Forward Waveform Modeling


Low frequencies (<1 Hz)

High frequencies (>1 Hz)

Deterministic character

Stochastic character

Numerical solutions of wave


equation (FEM, FDM, SEM)

Stochastic solutions for random


wave field (Point-source, finitefault)

Hercules (CMU- FEM code),


3DFD_VS (Moczo and Kristek)

SMSIM (Boore), FINSIM, EXSIM

(Atkinson, Beresnev,
Motazedian)

(2f ) 2

Aij ( f ) C M 0ij H ij

fRij

e Q ( f ) G ( Rij ) D( f ) e f 0


f (t )
0ij

Broadband motions (Entire frequency range


of interest!)
1 Hz is critical for engineering structures
13

Forward Waveform Modeling (Limitations)


Low frequencies (<1 Hz)
Need a well-defined and highresolution 3D velocity model of
the region of interest
Computing power
Seismic stations and much
data to validate

High frequencies (>1 Hz)


Need to know the regional
input empirical parameters (path,
duration, geometric spreading
etc)
Computing power
Seismic stations and much
data to validate
Amplifications! (Again at least
1D velocity models required!)

Lets look at what we have done (and what we could not


do) for Turkey
in terms of forward and inverse modeling
14

Ground Motion Simulations

Stochastic Point-Source Mo
(descriptions by Atkinson et al., 2009, BSSA)
The shear wave amplitude spectrum in frequency domain is
the product of filter functions representing the source,
propagation and site effects.

Acc( M 0 , R, f ) Source( M 0 , f )Path ( R, f )Site( f )

The stochastic point-source model assumes that


the earthquake source is concentrated at a point.
Acceleration time series generated at a site carry
both deterministic and random aspects of
ground-motion shaking.

Stochastic Point-Source
Modeling
(descriptions
byaspects
Atkinson
et al., by
2009,
The deterministic
are specified
the BSSA)
average Fourier spectrum, typically as a function of
magnitude and distance.
The stochastic aspects are treated by modeling the
motions as noise with the specified underlying
spectrum.
The point-source assumption is reasonable when
the source-to-site distance is much larger than the
source dimensions

Finite-Fault Source
Models
Fault is assummed to be a finite rectangular plane and divided into

subfaults.

Each subfault is assummed to be a point source with an 2 spectrum.

Ground motions from each subfault are summed with a time delay in order
to obtain the ground motion acceleration from the entire fault as rupture
starts from the hypocenter:
a(t)=aij(t+tij)

Corner Frequency of the ijth subfault at any time is a fuction of the total
number of ruptured subfaults at that time.

Finite-Fault Source Models


(Figure adapted from Hisada, 2008, Journal of
Seismology)

Finite-Fault Source
Models

Finite-Fault Source
Models

References:

Akkar, S., Bommer, J.J. (2007). Prediction of elastic displacement response spectra in Europe and the
Middle East, Earthquake Engng Struct. Dyn, 36, 1275-1301.
Beresnev, I., and G. M. Atkinson (1997). Modeling finite-fault radiation from the w n spectrum, Bull. Seism.
Soc. Am. 87, 67 - 84.
Boore, D. M. (1983). Stochastic simulation of high-frequency ground motions based on seismological models
of the radiated spectra, Bull. Seism. Soc. Am. 73, 18651894.
Boore, D. M. (1984). Use of seismoscope records to determine M L and peak velocities, Bull. seism. Soc.
Am., 74, 315324.
Gulkan, P., Kalkan, E. (2002). Attenuation modeling of recent earthquakes in Turkey, Journal of Seismology
6, 397-409.
Herrmann, R.B. (1985). An extension of Random Vibration Theory estimates of strong ground motion to large
earthquakes, Bull. Seis. Soc. Am. 75, 14471453.
Mavroeidis G. P., Papageorgiou A. S. (2003). A mathematical representation of near-fault ground motions,
Bull. Seism. Soc. Am. 93, 10991131.
Motazedian, D., and G. M. Atkinson (2005). Stochastic finite-fault modeling based on a Dynamic Corner
Frequency, Bull. Seism. Soc. Am. 95, 9951010.
Nakamura, Y. (1989). A method for dynamic characteristics estimation of subsurface using microtremor on
the ground surface, QR RTRI 30, 2533.
Ugurhan, B. and A. Askan (2009). Stochastic Strong Ground Motion Simulation of the 12 November 1999
Dzce (Turkey) Earthquake Using a Dynamic Corner Frequency Approach, Bull. Seism. Soc. Am. in
revision.
Ulusay, R., Tuncay, E., Sonmez H., Gokceoglu C. (2004). An attenuation based on Turkish strong motion
data and iso-acceleration map of Turkey, Engineering Geology 74, 265-291.
Umutlu, N., K. Koketsu, and C. Milkereit (2004). The rupture process during the 1999 Duzce, Turkey,
earthquake from joint inversion of teleseismic and strong-motion data, Tectonophysics 391, 315 324.

Sources: Fault Zones in Turkey


>90% of the total area is
located in seismically active
regions
> 95 % of its population being
under earthquake threat
Mostly reinforced concrete and
brick masonry structures
Utkucu et al., 2003, GJI

Rapid ground and structural


vulnerability assessment
techniques: deterministic and
stochastic approaches

Akyuz et al., 2002, BSSA


23

Region of Interest:

Duzce, Bolu, Kaynasli: Northwestern towns hit by two


Mw 7 earthquakes in less than 3 months! 1999 Kocaeli
and Duzce earthquakes
Active faults around industrial facilities in NW Turkey
Major structural damage and loss of lives
Finite-fault simulations
24

Northwest of NAFZ
No velocity model, not even for Istanbul that is
supposed to be worked on really hard by
geophysicists
Existing models are coarse (poorly resolved and
mostly in the deeper structure ~1km, 2 km etc)
Geotechnical engineers, on the other hand, are
mostly concerned with the topmost layers
We do not know what goes on between 30m and
1000 m most of the time! (where the most of the
seismic amplifications/ attenuations take place)
Do we have to drill down to kms in urban regions?
NO

25

Passive Seismic Experiments (Collaboration with


Prof. Asten)

Spatial Autocorrelation Technique: For a plane wave passing pairs of


stations, coherency averaged in azimuth , gives
(f) = e i k.r d = Jo (k.r)
where k = 2 f / C(f) (Aki, 1957,1965)

DZC, BOL stations (both


major alluvial basins)
1 hour records of noise
no real seismic source: just
passive seismic recordings
(Asten 2002; 2004)

Passive Seismic Experiments

In progress for
comparison with logs
and other methods

i o

i o

m
(

m
(

h
e

Amplifications from
different techniques

(Asten et al.,2014)

/ s

May 7th, 2010


27
s

/ s

Application: Stochastic Simulation of the


1999 Dzce Earthquake

May 7th, 2010

28

Input Parameters of the 1999 Dzce eq.


(Ugurhan and Askan, 2010, BSSA)

29

Validation at Strong Motion Stations

(Ugurhan

and Askan, 2010, BSSA)

30

Comparison of Synthetics with Regional and


Global GMPEs (Ugurhan and Askan, 2010, BSSA)

31

Correlation of Observed Damage with


Simulated Ground Motion Distribution

(Ugurhan

and Askan, 2010, BSSA)


simulated PGA
values are compared
with the observed
damage
damage defined in
terms of Mean
Damage Ratio (MDR)
Pk ( DS , I )

N ( DS , I )
N (I )

MDR( I ) Pk ( DS , I ) CDR ( DS )
DS

32

LAquila (Italy) Earthquake


(April 2009)
Ugurhan, Askan, Akinci, Malagnini (2012)

April 29th, 2011

33

LAquila (Italy) Earthquake


(April 2009)

34

LAquila (Italy) Earthquake


(April 2009)

35

Can we really link simulations with practice?


(Ugurhan, Erberik and Askan, 2011, April, BSSA)

Identification of
Ground Motions

Fragility
Information of the
Building Stock
None

Earthquake
Loss Estimation

Complete

36

Can we really link simulations with practice?


(Ugurhan, Erberik and Askan, 2011, April, BSSA)

37

Ground Motion Simulations (Turkey case)


Deterministic scenarios
Finite-fault simulations (EXSIM, Motazedian and
Atkinson, 2005)

A range of Mw= 5-7.5 on Duzce fault

Fault area (Wells and Coppersmith, 1994), stress drop


(Mohammadioun and Serva, 2001), random slip
distributions
Site amplifications are estimated from previous
data H/V ratios (a passive seismic testing project is
in process)

First validation for 1999 Duzce eq and then


scenarios!

38

Ground Motion Distributions (PGA) of Scenario


Earthquakes

May 7th, 2010

39

Fragility Analyses
reference fragility curves were derived for the specific regional
building types (Erberik, 2008, Eq. Engr. Struct. Dyn.)
R/C: low-mid rise, including degradation properties of concrete,
masonry: number of storeys, material type and quality, regularity
in plan, openings in walls
pushover analysis of typical buildings under a range of input
ground motions
validated using GMPEs for the major earthquakes in the region
(Erberik, 2008, Engineering Structures)
hazard parameter: PGV for R/C, PGA for masonry
None

Light
Moderate
Severe

Calculate the
probability of being
in each damage
state referring to the
hazard intensity

40

Fragility Analyses
For cities of Bolu, Duzce, and Kaynal:
For each magnitude Mw, PGA and PGV distributions
are estimated in cells
Building type distribution and numbers in the cells
are estimated
Each building type is associated with its fragility
Damage ratios are calculated for three Limit States
(N+L, M, S+C) from the syntetic PGA/PGV distributions
as the input ground motion

41

Can we really link simulations with practice?


YES
(Ugurhan, Erberik and Askan, 2011, April, BSSA)

May 7th, 2010

42

(Bridging) the gap between


seismology and engineering

Most of the loss estimation studies are either at


a global scale
Or contrarily focus on only a set of buildings
The detailed focus is either on site response or
ground motion modeling or building response

43

(Bridging) the gap between


seismology and engineering
Can we actually estimate seismic
losses in a city using only locallyderived components?

44

Objective
Regional seismic activity (Both Probabilistic and
Deterministic Framework for ground motions )
Local site conditions (Velocity models!)
Distinct properties of the building stock of interest
The results to be used
Hazard mitigation
Disaster management
Insurance premiums

45

Study Area: Erzincan

46

Task 1. Site Conditions


(Askan et al. 2015)

Microtremors: MMSPAC method (Asten, 2000; 2002)


Fast
Relatively cheaper
Can resolve deeper layers with the addition of HVSR
information

Maresca et al. 2006


47

Task 1. Site Conditions

48

Task 1. Site Conditions

49

Task 3. Deterministic
Ground Motion Scenarios
very sparsely monitored until recently
1992 (Mw=6.6) earthquake is recorded by only 3 stations
Two approaches: 1. Pure stochastic, 2. Hybrid simulations
Validation of this mainshock (Askan et al., 2013)
4

10

10

10

NS

-500
500

EW

-500
500

Synthetic

0
-500

10

30

50

70

Time (Sec)

90

Frequency (Hz)

Acceleration (cm/sec)

Acceleration (cm/sec)

10

500

100

10

10

NS

-100
100

EW

-100
100

Synthetic

0
10

20

30

Time (Sec)

40

Frequency (Hz)

10

-100

Frequency (Hz)

10

10

10

Frequency (Hz)
1

NS
EW
Synthetic

10

10

10

FAS (cm/sec)

NS
EW
Synthetic

10

Station TER

10

10

10

Acceleration (cm/sec)

FAS (cm/sec)

10

Station REF

NS
EW
Synthetic

FAS (cm/sec)

Station ERC

10

50

10

Frequency (Hz)

10

NS

-50
50

EW

-50
50

Synthetic

-50

10

Time (Sec)

20

50

Task 3. Deterministic
Ground Motion Scenarios
Mw
(0.33
Mw(6.6)
(6.6)SA
PGA
(g)s) (g)

Mw
(1.0(cm/s)
s) (g)
Mw (6.6)
(6.6) SA
PGV

51

Task 3. Deterministic
Ground Motion Scenarios
Mw (7.0) PGA (g)

Mw (7.0) PGV (cm/s)

52

Task 4. Building Fragility


Functions

On site structural classification


12 RC, 9 Masonry sub-classes
Dynamic properties of each sub-classes
(period, ductility, etc)
Monte-carlo sampling
Detailed nonlinear analyses within
OPENSEES platform

53

Task 4. Building Fragility


Functions

Ground motions are taken from the scenarios


in Task 3.
A set of motions including various Mw values,
at varying source-site distances and site
conditions
Objective is to encounter the local seismic
demand on buildings of interest

54

Generation of Fragility Curves


P [ LS | PGV ]

55

probability of exceedance

probability of exceedance

probability of exceedance

probability of exceedance

Task 4. Building Fragility


Functions

PGV (cm/s)

PGV (cm/s)
PGV (cm/s)

PGV (cm/s)
56

Damage Comparison:
1992 Event

57

Summary and What is

next?
Mathematical models are required to

define earthquake sources and wave


propagation mechanisms
Numerical solution of such models
and engineering knowledge can be
combined for practical applications

April 21, 2010


58

Summary and What is


Next:
next?

Loss estimations based on selected scenarios


Hybrid ground motion simulations
Integrated Earthquake Simulations (Citybuilding models) with Prof.Hori
Economic and social losses at city-level
Insurance premiums (Done! )

April 21, 2010


59

Colloborations and Projects


YESAP project by our team (2015-...) with Japanese
colleagues
TUJJB-UDP-01-12 Project by our team (2012-2015)
Tubitak 109M390: (Monash University) and Ankara
University, Geophysical Engineering, METU: passive
seismic tests in Duzce and Bolu for 1D velocity models
FP-7 NERA: Ground motion simulations
FP-7 SERIES: VCE, METU, AUTH for inversions in
Volvi region
EC: EMME Project: Several Middle Eastern countries
and institutions: simulations for GMPE scaling
Carnegie Mellon on the inverse modeling
2 small budget BAP projects: amplifications based

60

Thank you.
Aysegul Askan: aaskan@metu.edu.tr
http://blog.metu.edu.tr/aaskan

61

Das könnte Ihnen auch gefallen